Abstract
Background: Accurately predicting difficult airways is essential to ensuring patient safety in anesthesiology and emergency medicine. However, traditional assessment tools often lack sufficient sensitivity and specificity, particularly in high-pressure or resource-limited settings. Artificial intelligence (AI) and machine learning (ML) have emerged as promising tools for enhancing airway assessment. Objective: This review evaluates the performance of AI- and ML-based models for predicting difficult airways and compares them with traditional clinical methods. The review also analyzes the models' methodological robustness, clinical applicability, and ethical considerations. Methods: A comprehensive literature search was conducted across PubMed, Web of Science, and Scopus to identify studies published between 2020 and 2025 that employed AI/ML models to predict difficult airways. Both original research and review articles were included. Key metrics, such as the area under the curve (AUC), sensitivity, and specificity, were extracted and compared. A qualitative analysis was performed to focus on dataset characteristics, validation strategies, model interpretability, and clinical relevance. Results: AI models demonstrated superior performance compared to traditional assessment tools. The MixMatch semi-supervised deep learning (DL) model achieved the highest performance (area under the curve [AUC] of 0.9435, sensitivity of 89.58%, and specificity of 90.13%). Models that used facial imaging combined with deep learning consistently outperformed those that relied solely on clinical parameters. However, methodological heterogeneity, a lack of standardized evaluation metrics, and limited population diversity impeded cross-study comparability. Few studies incorporated interpretability frameworks or addressed ethical challenges related to data privacy and algorithmic bias. Conclusions: AI and ML models have the potential to transform the assessment of difficult airways by improving diagnostic accuracy and enabling real-time clinical decision support.